vulcand
add additional buffer limit test cases
When looking into https://github.com/traefik/traefik/issues/10687, I came across some interesting behavior where the buffer changes from returning a 200 to a 500 depending on the configured limits, but then also changes back and forth when crossing a size equal to a multiple of 32768 bytes.
For instance, the traefik issue configures a buffer that only sets the maxResponseBodyBytes option to a value of 2000 bytes. For response sizes from 1 up to 2000 bytes, the buffer behaves correctly and returns a 200 status which is expected. From 2001 up to 32768, the buffer returns a 500 status which is also expected. From 32769 up to 34769 (32769 + 2000) the buffer changes back to returning a 200 status, which I don't think is the expected behavior. From 34769 up to 65538 the behavior changes back to a 500, and so on at size boundaries equal to multiples of 32768.
This PR is not mean to be merged in its current state, but adds a handful of test cases that demonstrate the observed behavior. Based on my understanding of the expected behavior, each case should result in a 500 response, but a few of them result in unexpected 200 responses.
$ go test -v ./buffer -count=1 -run TestBuffer_responseLimitReached
=== RUN TestBuffer_responseLimitReached
=== RUN TestBuffer_responseLimitReached/small_limit_with_body_larger_than_max_response_bytes
=== RUN TestBuffer_responseLimitReached/small_limit_with_body_larger_than_32768_bytes
buffer_test.go:243:
Error Trace: /Users/adamd/go/src/github.com/adamvduke/oxy/buffer/buffer_test.go:243
Error: Not equal:
expected: 500
actual : 200
Test: TestBuffer_responseLimitReached/small_limit_with_body_larger_than_32768_bytes
=== RUN TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_bytes
buffer_test.go:243:
Error Trace: /Users/adamd/go/src/github.com/adamvduke/oxy/buffer/buffer_test.go:243
Error: Not equal:
expected: 500
actual : 200
Test: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_bytes
=== RUN TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_+_1999_bytes
buffer_test.go:243:
Error Trace: /Users/adamd/go/src/github.com/adamvduke/oxy/buffer/buffer_test.go:243
Error: Not equal:
expected: 500
actual : 200
Test: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_+_1999_bytes
=== RUN TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_+_2000_bytes
=== RUN TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_65536_+_1999_bytes
buffer_test.go:243:
Error Trace: /Users/adamd/go/src/github.com/adamvduke/oxy/buffer/buffer_test.go:243
Error: Not equal:
expected: 500
actual : 200
Test: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_65536_+_1999_bytes
=== RUN TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_65536_+_2000_bytes
--- FAIL: TestBuffer_responseLimitReached (0.01s)
--- PASS: TestBuffer_responseLimitReached/small_limit_with_body_larger_than_max_response_bytes (0.00s)
--- FAIL: TestBuffer_responseLimitReached/small_limit_with_body_larger_than_32768_bytes (0.00s)
--- FAIL: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_bytes (0.00s)
--- FAIL: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_+_1999_bytes (0.00s)
--- PASS: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_32768_+_2000_bytes (0.00s)
--- FAIL: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_65536_+_1999_bytes (0.00s)
--- PASS: TestBuffer_responseLimitReached/larger_limit_with_body_larger_than_65536_+_2000_bytes (0.00s)
FAIL
FAIL github.com/vulcand/oxy/v2/buffer 0.643s
FAIL
I'm not sure I have enough context to know that this approach is correct. It does seem like the error when writing to the buffer should propagate back to the client rather than logging it and moving on, and there is an existing pattern of using b.errHandler.ServeHTTP(w, req, err) to propagate other error conditions.
@ldez I tested the PR with a Traefik setup, and it works fine. Let me know if you need any specific feedback.